APEX²: Adaptive and Extreme Summarization for Personalized Knowledge Graphs

Figure credit: (one of) my favorite video game APEX Legends. This project is dedicated to the memory of my undergraduate years.

---

Paper (full arXiv version) | Paper (ACM version) | Poster | Video Summary

file structure

- README.md (this file)
- .gitignore
- code
    - src (utils)
    - main.py (run this, specific commands will be provided below)
    
- PKG_EXP
    - DBpedia
        - queries (query folder, empty, will be generated)
            - user # (user folder)
        - facts.gz (dataset, download instructions below)
    - MetaQA    
        kb.txt (dataset)
        - dataprocessing.py (utils when generating queries)
        - generate_query.py (can be used to re-generate queries)
        - queries (query folder)
            - user # (user folder)
    - YAGO
        - queries (query folder, empty, will be generated)
            - user # (user folder)
        - yagoFacts.gz, missing large dataset file, need to be manually downloaded. Instructions in both paper and below, 

Datasets

We provide MetaQA that we used in our experiments along with the source code.

DBpedia can be download from http://downloads.dbpedia.org/3.5.1/en/. Specifically, download the "mappingbased_properties_en.nt" file. Rename it to 'facts.gz' and place it as PKG_EXP/DBpedia/facts.gz.

YAGO can downloaded from https://www.mpi-inf.mpg.de/departments/databases-and-information-systems/research/yago-naga/yago/downloads/. The button is Download YAGO Themes -> CORE -> yagoFacts -> Download TSV. If clicking on "Download TSV" doesn't work, right click "Download TSV" and copy link address then open in a new window in your browser. After downloading, convert to ".gz" file using gzip if not in this format.

Commands to prepare the datasets:

# DBpedia
wget https://downloads.dbpedia.org/3.5.1/en/mappingbased_properties_en.nt.bz2
mv mappingbased_properties_en.nt.bz2 PKG_EXP/DBpedia/facts.gz
# YAGO
wget http://resources.mpi-inf.mpg.de/yago-naga/yago3.1/yagoFacts.tsv.7z
# if 7z is not installed, you can manually extract the 7z file on Windows/Mac machine
7z x yagoFacts.tsv.7z
gzip yagoFacts.tsv

How to run

Go to code/src/path.py, change the dataset path if needed. The resulted log file name can be set in main.py line 14.

For each dataset, you can run the corresponding command. We recommand to run MetaQA first because DBpedia and YAGO are very large knowledge graph datasets and takes time to finish executing.

cd code

# quick start with pre-generated query on MetaQA
python main.py --kg MetaQA --method apex apex-n --percent-triples 0.0001 --n_users 1

# generate user queries and test on DBpedia and YAGO
python main.py --kg DBpedia --method apex apex-n --save-queries --percent-triples 0.000001 --n_users 1

python main.py --kg YAGO --method apex apex-n --save-queries --percent-triples 0.000001 --n_users 1

If you want to use the same queries stored, change --save-queries to --load-queries. When run with --save-queries, the query files will be re-written, you need to enter "c" when pdb askes whether to proceed.

Result

For convenience, we save the detailed experiment results in the log files. We record F1 score and running time for each adapting. To quickly view the result, search "Ave Time on Each Training Log" and "Ave Ave F1 on Each Training Log" in the log files. Note that the "Summarizing with ......" under "Ave Ave F1 on Each Training Log" is not a pair. It's the next method applied after the method that generated this result.

Cite

If you find this repository useful in your research, please consider citing the following paper:

@inproceedings{DBLP:conf/kdd/LiFAH25,
  author       = {Zihao Li and
                  Dongqi Fu and
                  Mengting Ai and
                  Jingrui He},
  editor       = {Yizhou Sun and
                  Flavio Chierichetti and
                  Hady W. Lauw and
                  Claudia Perlich and
                  Wee Hyong Tok and
                  Andrew Tomkins},
  title        = {APEX\({}^{\mbox{2}}\): Adaptive and Extreme Summarization for Personalized
                  Knowledge Graphs},
  booktitle    = {Proceedings of the 31st {ACM} {SIGKDD} Conference on Knowledge Discovery
                  and Data Mining, V.1, {KDD} 2025, Toronto, ON, Canada, August 3-7,
                  2025},
  pages        = {741--752},
  publisher    = {{ACM}},
  year         = {2025},
  url          = {https://doi.org/10.1145/3690624.3709213},
  doi          = {10.1145/3690624.3709213},
  timestamp    = {Tue, 08 Jul 2025 09:19:45 +0200},
  biburl       = {https://dblp.org/rec/conf/kdd/LiFAH25.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}